Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Canada → A machine learning approach to understanding the effects of climate change and competition on growth of coihue (Nothofagus dombeyi) and lenga (Nothofagus pumilio) species in northern Patagonia, Argentina

Galvão de Sousa Magalhães, Juliana (2020)

A machine learning approach to understanding the effects of climate change and competition on growth of coihue (Nothofagus dombeyi) and lenga (Nothofagus pumilio) species in northern Patagonia, Argentina

Galvão de Sousa Magalhães, Juliana

Titre : A machine learning approach to understanding the effects of climate change and competition on growth of coihue (Nothofagus dombeyi) and lenga (Nothofagus pumilio) species in northern Patagonia, Argentina

Auteur : Galvão de Sousa Magalhães, Juliana

Université de soutenance : University of British Columbia

Grade : DOCTOR OF PHILOSOPHY (PhD) 2020

Résumé
The ongoing process of a warming climate is impacting forests. A warmer climate scenario alters the availability of growth resources. The more resource availability decreases, the more trees compete, and that additional stress effect on growth pushes trees usually to their survival limits. Competition might act in combination with climate in exacerbating the impacts of climate change on forests. The fact that competition interacts with climate, site and other factors complicate simulation of tree growth responses to climate change with a traditional modeling method. Nevertheless, advances in machine learning algorithms have brought tools with scientific applications, such as algorithmic modeling. Unlike traditional statistical modeling, ML algorithms are more flexible in dealing with nonlinear relationships between variables. In this study, the concept of algorithmic modeling was used to investigate the role of competition on tree growth responses to climate change. Recurrent Neural Network (RNN) algorithms have an upgraded architecture that can retain past information, which helps capture tree growth unidirectional flow. In this regard, I designed the Tree Growth Prediction App to introduce the RNN algorithm as a new modeling technique. To test the App, I decided to study coihue (Nothofagus dombeyi) and lenga (Nothofagus pumilio) growth responses in northern Patagonia, where a warmer climate scenario is being predicted. Results from the RNN tree growth model indicate that the size-asymmetric competition effect is more important than the climate effect on growth. N. dombeyi, as the larger individual species, has a competitive advantage over N. pumilio in wet sites, but N. pumilio might withstand water competition as a result of a shallow root system that can pre-empt water supplies from N. dombeyi with deeper roots. Overall, the results of competition importance to the climate effect on tree growth are essential information to more realistically project forest responses to climate change. Furthermore, the Tree Growth Prediction App is a methodological contribution to a current generation of advanced models that help researchers better understand tree growth responses to climate change.

Présentation et version intégrale

Page publiée le 13 avril 2021